[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"globals":3,"page:\u002Fresources\u002Fdatabricks\u002Fdatabricks-ai-extract-precision-mode:published":576},{"articleCategories":4,"databricksService":66,"defaultSeo":72,"footer":132,"header":256,"pageSlugsByUid":303,"settings":314},[5,19,31,43,54],{"_createdAt":6,"_id":7,"_rev":8,"_system":9,"_type":12,"_updatedAt":13,"name":14,"orderRank":15,"slug":16,"uid":18},"2025-12-02T14:52:22Z","3b9cf7fe-e7ca-440e-99f6-0b13db9e4d30","TITHlt8XkiMnotzVXfF9P0",{"base":10},{"id":7,"rev":11},"P6h3aQpzv40g9XSVyQ7aI5","articleCategory","2026-01-20T14:41:53Z","Databricks","0|100008:",{"_type":17,"current":18},"gluharSlug","databricks",{"_createdAt":20,"_id":21,"_rev":22,"_system":23,"_type":12,"_updatedAt":26,"name":27,"orderRank":28,"slug":29,"uid":30},"2025-12-02T14:52:59Z","479bcd34-5ae8-4742-a1c1-22de4aa1a3ba","TITHlt8XkiMnotzVXfF9XQ",{"base":24},{"id":21,"rev":25},"5FspTM65O6BRJGwT4eIvSq","2026-01-20T12:41:04Z","Data Strategy","0|10000o:",{"_type":17,"current":30},"data-strategy",{"_createdAt":32,"_id":33,"_rev":34,"_system":35,"_type":12,"_updatedAt":38,"name":39,"orderRank":40,"slug":41,"uid":42},"2025-12-02T14:53:27Z","d1601cf0-158c-4be9-bdeb-95af4c13d719","TITHlt8XkiMnotzVXfF9TD",{"base":36},{"id":33,"rev":37},"d1kEZFX7awui5VEkfbEeqX","2026-01-20T12:41:16Z","Technical Guides","0|100014:",{"_type":17,"current":42},"technical-guides",{"_createdAt":44,"_id":45,"_rev":34,"_system":46,"_type":12,"_updatedAt":49,"name":50,"orderRank":51,"slug":52,"uid":53},"2025-12-02T14:54:55Z","88763297-bf8b-426c-a8af-f1d9e88d971e",{"base":47},{"id":45,"rev":48},"yLjzPDnEPDnZbsBE0m74iT","2026-01-20T12:40:27Z","Industry Solutions","0|10002w:",{"_type":17,"current":53},"industry-solutions",{"_createdAt":55,"_id":56,"_rev":57,"_system":58,"_type":12,"_updatedAt":61,"name":62,"orderRank":63,"slug":64,"uid":65},"2025-12-02T14:55:18Z","b07b3aee-dbbf-4bbf-9227-27f4c8ae29c8","4Hcc1NJ6s8BorOrDxp9fhs",{"base":59},{"id":56,"rev":60},"yLjzPDnEPDnZbsBE0m71fA","2026-01-20T12:39:44Z","Company News","0|10003c:",{"_type":17,"current":65},"company-news",{"_id":67,"_type":68,"description":69,"name":14,"slug":70,"uid":18},"772fd7d5-97fb-4525-9e9c-13279a8c3efb","service","Lakehouse Platform Implementation",{"_type":17,"current":71},"databricks-consulting-implementation-services",{"_createdAt":73,"_id":74,"_rev":75,"_system":76,"_type":74,"_updatedAt":79,"description":80,"ogImage":81,"title":131},"2026-01-14T11:07:28Z","defaultSeo","ABl7P4fLW27D0dQw5Aguqg",{"base":77},{"id":74,"rev":78},"Coi90aDgVBbfPSN8whzpsv","2026-08-07T15:11:57Z","Databricks Select Consulting Partner delivering Lakehouse implementation, migration, governance, and production AI to EMEA enterprises. 90-day delivery. ISO 27001 certified. 300Euro to 600 Euro a day.",{"_type":82,"asset":83},"image",{"_createdAt":84,"_id":85,"_rev":86,"_type":87,"_updatedAt":88,"assetId":89,"extension":90,"metadata":91,"mimeType":125,"originalFilename":126,"path":127,"sha1hash":89,"size":128,"uploadId":129,"url":130},"2026-08-07T15:11:05Z","image-f41d0f78de6cd468de22a82566c9c876cc069b63-1921x1077-png","ABl7P4fLW27D0dQw5AgRME","sanity.imageAsset","2026-08-07T15:11:06Z","f41d0f78de6cd468de22a82566c9c876cc069b63","png",{"_type":92,"blurHash":93,"dimensions":94,"hasAlpha":99,"isOpaque":100,"lqip":101,"palette":102,"thumbHash":124},"sanity.imageMetadata","MBCY{:t69bfPoJ_N$%xabHayENofxas:of",{"_type":95,"aspectRatio":96,"height":97,"width":98},"sanity.imageDimensions",1.7836583101207057,1077,1921,false,true,"data:image\u002Fpng;base64,iVBORw0KGgoAAAANSUhEUgAAABQAAAALCAIAAADwazoUAAAACXBIWXMAAAsTAAALEwEAmpwYAAACWUlEQVR4nBXG+U\u002FaUAAA4P4TS5bpBogth3KI4\u002FBAEyaeYw4VIULvvvb19fW1pYelHHIYl5hsi\u002FuLF5Pvh496Dt1V6C5DdxG688fhUzCc+cOZ70x9Z+za4dAOHNu3Lde2XEIc07RMkxjYRNiAiFoGziKw54H95Nszz5569sS1Jq4186yZSyKHeBZxTeyY2MaGhRDWDaQjXUOaCqm5S56G5nRoTpx3Y8eMbHNkGRPHWHl44RoBQTZCRIcmhIam6UDVFBXIQJEUygFigEBEYERQRNCIoNDUAxM+Yjh34C9PHxNoaSpWAQIAKrImyYogyYIkcgJ1c9GU7tu+JkSGOjJAaIBHBAIEfB2MsfLigJmp2KpsyJIuiZogAp6XOV5kOX7AUofl0vXJMR50IiiMdXGsi5EuRLoYwvcssDAzeBfwROawyCKe1di+0u9JD12xd09VdoqNWrnfOrXZ25HSm8H+ArELg1sY\u002FBLzzya3wtwEcqE68KSew3Vwv612WvLtpdQ+p0q5fCVfaNQqd80j9ebMY39MlbuV3nu12DeX\u002F+fzf4f8Kxkstc6IbzndM\u002FizIVzUH5p7vW9VKkOntzdTxVR2L58\u002Fre52GvtSq0G6F3Ol\u002FYd036zub+P2Rb2esGekfSScVjv10vdq4fxrrlnapmJr8cR6nI4ls8lUMZWuZrPHpULroNw\u002FOVAuj9Srunp1CK4OuWatvV882dmqb6VraabK0BWGoT59WFv\u002F+PnLenwjvskkaCaWzGzQRZopb9LlZLJM03vZzHExXy\u002Fkqpn0LsPsJOl8PJGLxbfjif9ub+rOY2uKowAAAABJRU5ErkJggg==",{"_type":103,"darkMuted":104,"darkVibrant":109,"dominant":112,"lightMuted":113,"lightVibrant":116,"muted":119,"vibrant":122},"sanity.imagePalette",{"_type":105,"background":106,"foreground":107,"population":108,"title":107},"sanity.imagePaletteSwatch","#523b30","#fff",3.96,{"_type":105,"background":110,"foreground":107,"population":111,"title":107},"#4c2c24",0.05,{"_type":105,"background":106,"foreground":107,"population":108,"title":107},{"_type":105,"background":114,"foreground":107,"population":115,"title":107},"#673c31",0,{"_type":105,"background":117,"foreground":118,"population":115,"title":107},"#d4aea5","#000",{"_type":105,"background":120,"foreground":107,"population":121,"title":107},"#a27f64",2.78,{"_type":105,"background":123,"foreground":107,"population":115,"title":107},"#ad6451","VhgKDIJHd3efeYeFiHoHwnECPA==","image\u002Fpng","home.png","images\u002Fwn2c54m4\u002Fv2\u002Ff41d0f78de6cd468de22a82566c9c876cc069b63-1921x1077.png",990661,"jJqWWPThhjYzs77pnTlHaGnDuoEllYXl","https:\u002F\u002Fcdn.sanity.io\u002Fimages\u002Fwn2c54m4\u002Fv2\u002Ff41d0f78de6cd468de22a82566c9c876cc069b63-1921x1077.png","Databricks Consulting Partner in EMEA | Cosmos Thrace",{"_createdAt":133,"_id":134,"_rev":135,"_system":136,"_type":134,"_updatedAt":139,"badges":140,"contacts":164,"leftColumn":180,"rightColumn":236,"slogan":255},"2025-12-05T08:17:46Z","footer","TITHlt8XkiMnotzVXfGEPx",{"base":137},{"id":134,"rev":138},"7wJzgGa1ow12Mnj8KY5s9e","2026-08-07T15:11:59Z",[141,153],{"_key":142,"_type":143,"alt":144,"asset":145},"20ac8c475e0b","imageWithAlt","ISO 9001",{"_id":146,"_type":87,"metadata":147,"url":152},"image-1620cd48eefa59d8cadce066bc2552204e7271f2-1146x1148-png",{"dimensions":148},{"_type":95,"aspectRatio":149,"height":150,"width":151},0.9982578397212544,1148,1146,"https:\u002F\u002Fcdn.sanity.io\u002Fimages\u002Fwn2c54m4\u002Fv2\u002F1620cd48eefa59d8cadce066bc2552204e7271f2-1146x1148.png",{"_key":154,"_type":143,"alt":155,"asset":156},"87217df5f868","ISO 27001",{"_id":157,"_type":87,"metadata":158,"url":163},"image-6d90ea8fdc162931376a72236a557357b58500d4-477x472-png",{"dimensions":159},{"_type":95,"aspectRatio":160,"height":161,"width":162},1.0105932203389831,472,477,"https:\u002F\u002Fcdn.sanity.io\u002Fimages\u002Fwn2c54m4\u002Fv2\u002F6d90ea8fdc162931376a72236a557357b58500d4-477x472.png",[165,169,173,176],{"_key":166,"label":167,"url":168},"1e670981db3e","hi@cosmosthrace.com","email",{"_key":170,"label":171,"url":172},"7f35924204bc","+359 876 102 110","phone",{"_key":174,"label":175,"url":172},"4e77273819c3","+44 7565 457 651",{"_key":177,"label":178,"url":179},"2be89609c74e","Sofia Tech Park,\u2028\nIncubator Building, \n111 B Tsarigradsko Shosse Blvd., \n1784 Sofia, Bulgaria","https:\u002F\u002Fmaps.app.goo.gl\u002FMtFzaGj5CYGW4C4t6",{"groups":181},[182,210],{"links":183,"title":209},[184,191,200],{"_key":185,"_type":186,"customName":100,"customUrl":99,"name":187,"newTab":99,"page":188},"a52cbde3587e","link","Databricks Implementation Services",{"_id":67,"_type":68,"category":189,"description":69,"name":14,"slug":190,"uid":18},null,{"_type":17,"current":71},{"_key":192,"_type":186,"customName":100,"customUrl":99,"name":193,"newTab":99,"page":194},"5005c47619ca","Data Strategy & Migration",{"_id":195,"_type":68,"category":189,"description":196,"name":197,"slug":198,"uid":199},"22368305-ed07-4d8c-8c7a-b453db581571","Enterprise Data Modernization","Data",{"_type":17,"current":199},"data",{"_key":201,"_type":186,"customName":100,"customUrl":99,"name":202,"newTab":99,"page":203},"f2b1f5946d2f","AI Data Solutions",{"_id":204,"_type":68,"category":189,"description":205,"name":206,"slug":207,"uid":208},"d7e33cae-48b3-4e03-afd1-0b7d8879a44e","Production AI Development","AI",{"_type":17,"current":208},"ai","Services",{"links":211,"title":235},[212,221,228],{"_key":213,"_type":186,"customName":99,"customUrl":99,"newTab":99,"page":214},"ed9ed7964670",{"_id":215,"_type":216,"category":189,"description":189,"name":217,"slug":218,"uid":220},"53c1b689-06fd-4f5c-aaae-30910debe72d","page","About Us",{"_type":17,"current":219},"about-us","about",{"_key":222,"_type":186,"customName":99,"customUrl":99,"newTab":99,"page":223},"998d04a94290",{"_id":224,"_type":216,"category":189,"description":189,"name":225,"slug":226,"uid":227},"277a5c88-4b91-44b8-b7be-f3410af4e4d4","Testimonials",{"_type":17,"current":227},"testimonials",{"_key":229,"_type":186,"customName":99,"customUrl":99,"newTab":99,"page":230},"2e07f8d2e934",{"_id":231,"_type":216,"category":189,"description":189,"name":232,"slug":233,"uid":234},"2e18054b-3dec-4e2b-8a7c-7f654d495f0a","Case Studies",{"_type":17,"current":234},"case-studies","Links",{"groups":237},[238],{"links":239,"title":254},[240,247],{"_key":241,"_type":186,"customName":99,"customUrl":99,"newTab":99,"page":242},"cb143f65e4be",{"_id":243,"_type":216,"category":189,"description":189,"name":244,"slug":245,"uid":246},"9bdb02e9-cb0e-4814-b16b-761351559df2","Contact Us",{"_type":17,"current":246},"contact-us",{"_key":248,"_type":186,"customName":99,"customUrl":99,"newTab":99,"page":249},"816051efcec1",{"_id":250,"_type":216,"category":189,"description":189,"name":251,"slug":252,"uid":189},"c9efda41-eb50-4040-89bb-da3a184f684a","Privacy Policy",{"_type":17,"current":253},"privacy-policy","Help","We go the extra mile with an extra smile, until your Databricks platform pays for itself, quarter over quarter. \nCustomers are always happy about our big 4 outcomes: \nNearshore prices, No pitch deck, no theatre, no surprise invoices.\nTalk to us before the next quarter slips.\n\n",{"_createdAt":257,"_id":258,"_rev":135,"_system":259,"_type":258,"_updatedAt":139,"button":262,"cornerText":265,"links":266},"2025-12-03T22:11:31Z","header",{"base":260},{"id":258,"rev":261},"IWGg7SNw7BhzqgYuATsGnE",{"_type":186,"page":263},{"_id":243,"_type":216,"category":189,"description":189,"name":244,"slug":264,"uid":246},{"_type":17,"current":246},"Databricks Silver Partner",[267,288,292,296],{"_key":268,"_type":269,"customName":100,"customUrl":99,"dropdown":270,"name":209,"newTab":99,"page":286},"1fc93c918075","linkWithDropdown",[271,276,281],{"_key":272,"_type":186,"customName":100,"customUrl":99,"name":273,"newTab":99,"page":274},"5e5c0f4b3809","Databricks Services",{"_id":67,"_type":68,"category":189,"description":69,"name":14,"slug":275,"uid":18},{"_type":17,"current":71},{"_key":277,"_type":186,"customName":100,"customUrl":99,"name":278,"newTab":99,"page":279},"f95f35556b5a","Data Services",{"_id":195,"_type":68,"category":189,"description":196,"name":197,"slug":280,"uid":199},{"_type":17,"current":199},{"_key":282,"_type":186,"customName":100,"customUrl":99,"name":283,"newTab":99,"page":284},"f9b35da3369b","AI Solutions",{"_id":204,"_type":68,"category":189,"description":205,"name":206,"slug":285,"uid":208},{"_type":17,"current":208},{"_id":67,"_type":68,"category":189,"description":69,"name":14,"slug":287,"uid":18},{"_type":17,"current":71},{"_key":289,"_type":269,"customName":99,"customUrl":99,"dropdown":189,"newTab":99,"page":290},"67c60109dd1b",{"_id":231,"_type":216,"category":189,"description":189,"name":232,"slug":291,"uid":234},{"_type":17,"current":234},{"_key":293,"_type":269,"customName":99,"customUrl":99,"dropdown":189,"newTab":99,"page":294},"52084a5690c8",{"_id":215,"_type":216,"category":189,"description":189,"name":217,"slug":295,"uid":220},{"_type":17,"current":219},{"_key":297,"_type":269,"customName":99,"customUrl":99,"dropdown":189,"newTab":99,"page":298},"f434c0657262",{"_id":299,"_type":216,"category":189,"description":189,"name":300,"slug":301,"uid":302},"e9c5c1cb-7ba7-4ff2-b55e-c63cc8d1e655","Resources",{"_type":17,"current":302},"resources",[304,305,306,307,309,310,311],{"slug":227,"uid":227},{"slug":234,"uid":234},{"slug":219,"uid":220},{"slug":308,"uid":308},"careers",{"slug":246,"uid":246},{"slug":302,"uid":302},{"slug":312,"uid":313},"\u002F","homepage",{"_createdAt":315,"_id":316,"_rev":317,"_system":318,"_type":316,"_updatedAt":321,"applyFormRecipient":322,"contactDetails":323,"contactFormConsent":334,"contactFormRecipient":356,"cookies":357,"oldSlugs":360},"2025-12-01T14:17:44Z","settings","47Kd9AFsDDpPkhM0B7w7jF",{"base":319},{"id":316,"rev":320},"ABl7P4fLW27D0dQw5Agz1w","2026-08-11T13:09:07Z","idan@cosmosthrace.com",[324,327,329,331],{"_key":325,"label":326,"url":172},"d57c36731170","+359 876102110",{"_key":328,"label":175,"url":172},"a659f0701cbd",{"_key":330,"label":167,"url":168},"3e17015a959b",{"_key":332,"label":333,"url":179},"313ffa7b4827","Sofia Tech Park, Incubator Building, 111 B Tsarigradsko Shosse Blvd., 1784 Sofia, Bulgaria",[335],{"_key":336,"_type":337,"children":338,"markDefs":352,"style":355},"9ead364e59eb","block",[339,344,348],{"_key":340,"_type":341,"marks":342,"text":343},"750ca13120df","span",[],"By submitting this form, we have your consent that Cosmos Thrace may process your personal data to respond to your inquiry. You can read more in our ",{"_key":345,"_type":341,"marks":346,"text":251},"ad7f9f61ef7a",[347],"fa2e0964302b",{"_key":349,"_type":341,"marks":350,"text":351},"e2433a35e963",[],".",[353],{"_key":347,"_type":186,"href":354},"\u002Fprivacy-policy","normal","hi@cosmosthrace.com, mladen@cosmosthrace.com, idan@cosmosthrace.com",{"buttonText":358,"description":359},"Got it.","We use a small number of cookies in order to make your experience more enjoyable and optimise performance.",[361,368,373,378,383,388,393,398,403,408,413,419,425,431,437,443,448,454,460,466,471,477,483,489,495,501,507,513,519,525,531,537,543,549,554,560,566,571],{"_key":362,"oldSlug":363,"redirect":364},"0164fcf0341b","our-cosmos-data-management-careers",{"_type":186,"customName":99,"customUrl":99,"newTab":99,"page":365,"query":367},{"_ref":299,"_type":366},"reference","category=our-stories",{"_key":369,"oldSlug":370,"redirect":371},"54ccdc393f04","bi-services",{"_type":186,"customName":99,"customUrl":99,"newTab":99,"page":372},{"_ref":195,"_type":366},{"_key":374,"oldSlug":375,"redirect":376},"2dc7b39a7a74","our-data-and-ai-services",{"_type":186,"customName":99,"customUrl":99,"newTab":99,"page":377},{"_ref":204,"_type":366},{"_key":379,"oldSlug":380,"redirect":381},"cf66109eda04","form-a-data-and-ai-partnership",{"_type":186,"customName":99,"customUrl":99,"newTab":99,"page":382},{"_ref":243,"_type":366},{"_key":384,"oldSlug":385,"redirect":386},"3788ce0e045a","data-and-ai-infographics",{"_type":186,"customName":99,"customUrl":99,"newTab":99,"page":387},{"_ref":299,"_type":366},{"_key":389,"oldSlug":390,"redirect":391},"70b59e49607e","data-and-ai-news",{"_type":186,"customName":99,"customUrl":99,"newTab":99,"page":392},{"_ref":299,"_type":366},{"_key":394,"oldSlug":395,"redirect":396},"ed4ffec9cb62","databricks-partnership",{"_type":186,"customName":99,"customUrl":99,"newTab":99,"page":397},{"_ref":67,"_type":366},{"_key":399,"oldSlug":400,"redirect":401},"3a5cd12a0576","databricks-consulting",{"_type":186,"customName":99,"customUrl":99,"newTab":99,"page":402},{"_ref":7,"_type":366},{"_key":404,"oldSlug":405,"redirect":406},"bbdda5690251","home",{"_type":186,"customName":99,"customUrl":99,"newTab":99,"page":407},{"_ref":313,"_type":366},{"_key":409,"oldSlug":410,"redirect":411},"b2e13b3c7613","frequently-asked-questions",{"_type":186,"customName":99,"customUrl":99,"newTab":99,"page":412},{"_ref":67,"_type":366},{"_key":414,"oldSlug":415,"redirect":416},"373453c655da","dax-in-power-bi",{"_type":186,"customName":99,"customUrl":99,"newTab":99,"page":417},{"_ref":418,"_type":366},"0e84c36f-2c69-4ed5-804b-1048f69bf260",{"_key":420,"oldSlug":421,"redirect":422},"0876bcb20800","examples-of-data-analytics-in-retail",{"_type":186,"customName":99,"customUrl":99,"newTab":99,"page":423},{"_ref":424,"_type":366},"9d1b76b7-7fd3-4655-9fb4-f64342167768",{"_key":426,"oldSlug":427,"redirect":428},"blogfix-01","resources\u002Fblog\u002Fautumn-basscom-retreat",{"_type":186,"page":429},{"_ref":430,"_type":366},"70b74696-dc20-4d33-beba-f55fd70edc9b",{"_key":432,"oldSlug":433,"redirect":434},"blogfix-02","resources\u002Fblog\u002Fbulgaria-took-the-first-step-toward-becoming-the-leader-in-eastern-europe-in-artificial-intelligence-and-computer-science",{"_type":186,"page":435},{"_ref":436,"_type":366},"990e3150-6ca9-43fb-8fc6-a99499850701",{"_key":438,"oldSlug":439,"redirect":440},"blogfix-03","resources\u002Fblog\u002Fcosmos-thrace-achieves-databricks-consulting-partner-select-status",{"_type":186,"page":441},{"_ref":442,"_type":366},"27a1a9ce-05e0-474b-81ba-8d9e37ab7e45",{"_key":444,"oldSlug":445,"redirect":446},"blogfix-04","resources\u002Fblog\u002Fcosmos-thrace-achieves-databricks-select-consulting-partner",{"_type":186,"page":447},{"_ref":442,"_type":366},{"_key":449,"oldSlug":450,"redirect":451},"blogfix-05","resources\u002Fblog\u002Fcosmos-thrace-iso-27001-certification",{"_type":186,"page":452},{"_ref":453,"_type":366},"077a683e-f9b6-4788-af1f-1e7fe3636ab0",{"_key":455,"oldSlug":456,"redirect":457},"blogfix-06","resources\u002Fblog\u002Fdatabricks-and-openai-partnership-transforms-enterprise-ai-agent-development",{"_type":186,"page":458},{"_ref":459,"_type":366},"40923b4f-514a-42b6-ba4a-1860b3620dd6",{"_key":461,"oldSlug":462,"redirect":463},"blogfix-07","resources\u002Fblog\u002Fdatabricks-data-ai-summit-2025-recap-4-things-that-matter",{"_type":186,"page":464},{"_ref":465,"_type":366},"2ffcff12-2929-4736-810f-65177331a33d",{"_key":467,"oldSlug":468,"redirect":469},"blogfix-08","resources\u002Fblog\u002F4-big-moves-from-databricks-at-the-2025-data-ai-summit",{"_type":186,"page":470},{"_ref":465,"_type":366},{"_key":472,"oldSlug":473,"redirect":474},"blogfix-09","resources\u002Fblog\u002Ffrom-attendee-to-sponsor-what-we-learned-at-databricks-data-ai-amsterdam",{"_type":186,"page":475},{"_ref":476,"_type":366},"2aa42dd3-99a9-4e92-836b-fdc8823370d6",{"_key":478,"oldSlug":479,"redirect":480},"blogfix-10","resources\u002Fblog\u002Fidan-harel-joins-cosmos-thrace",{"_type":186,"page":481},{"_ref":482,"_type":366},"a9846f3b-ee0c-400c-a1c4-4d238a100ccc",{"_key":484,"oldSlug":485,"redirect":486},"blogfix-11","resources\u002Fblog\u002Frobust-automatic-bulgarian-speech-recognition-with-insaits-bggpt",{"_type":186,"page":487},{"_ref":488,"_type":366},"52155075-0cb6-479f-bc34-1ecc50a23789",{"_key":490,"oldSlug":491,"redirect":492},"blogfix-12","resources\u002Fblog\u002Fstart-of-our-horizon-journey",{"_type":186,"page":493},{"_ref":494,"_type":366},"c6b4f8d3-9408-41e8-95ef-f5a611f59dff",{"_key":496,"oldSlug":497,"redirect":498},"blogfix-13","resources\u002Fblog\u002Fmodern-data-governance-for-ai",{"_type":186,"page":499},{"_ref":500,"_type":366},"24c80ace-d32d-4eaa-9ea3-c3b63f7512b9",{"_key":502,"oldSlug":503,"redirect":504},"blogfix-14","resources\u002Fblog\u002Fdigital-twins-model-in-construction",{"_type":186,"page":505},{"_ref":506,"_type":366},"f43a904d-2300-4905-ac3a-a920dc8de678",{"_key":508,"oldSlug":509,"redirect":510},"blogfix-15","resources\u002Fblog\u002Fhow-databricks-latest-updates-are-revolutionizing-drug-discovery-and-patient-outcomes",{"_type":186,"page":511},{"_ref":512,"_type":366},"2c98f82f-1502-4c99-9aa0-5ac60cca045d",{"_key":514,"oldSlug":515,"redirect":516},"blogfix-16","resources\u002Fblog\u002Fbi-reporting-analyst",{"_type":186,"page":517},{"_ref":518,"_type":366},"8b16ac0b-ac5f-4473-88a3-93cd9a51f179",{"_key":520,"oldSlug":521,"redirect":522},"blogfix-17","resources\u002Fblog\u002Fbi-reporting-business-requirements",{"_type":186,"page":523},{"_ref":524,"_type":366},"79e02131-66a4-4e53-9291-bbf02c8d3dbf",{"_key":526,"oldSlug":527,"redirect":528},"blogfix-18","resources\u002Fblog\u002Fdifference-between-bi-and-reporting",{"_type":186,"page":529},{"_ref":530,"_type":366},"ade89482-69f1-4ca6-a9cd-c24fbe101f49",{"_key":532,"oldSlug":533,"redirect":534},"blogfix-19","resources\u002Fblog\u002Fkey-terms-in-bi-reporting",{"_type":186,"page":535},{"_ref":536,"_type":366},"c50f94b4-2972-4140-bbc6-e4825995e36a",{"_key":538,"oldSlug":539,"redirect":540},"blogfix-20","resources\u002Fblog\u002Fpower-bi-adoption-roadmap",{"_type":186,"page":541},{"_ref":542,"_type":366},"pbiadopt2026-article",{"_key":544,"oldSlug":545,"redirect":546},"blogfix-21","resources\u002Fblog\u002Fwhat-is-business-intelligence-reporting",{"_type":186,"page":547},{"_ref":548,"_type":366},"0835e660-1a7a-4459-a7b7-8ae3430b062f",{"_key":550,"oldSlug":551,"redirect":552},"blogfix-22","resources\u002Fblog\u002Fbusiness-bi-reporting",{"_type":186,"page":553},{"_ref":548,"_type":366},{"_key":555,"oldSlug":556,"redirect":557},"blogfix-23","resources\u002Fblog\u002Fwhy-is-my-power-bi-report-not-refreshing",{"_type":186,"page":558},{"_ref":559,"_type":366},"pbiref2026-article",{"_key":561,"_type":562,"oldSlug":563,"redirect":564},"clientfix-01","object","the-different-job-roles-in-bi-reporting",{"_type":186,"page":565},{"_ref":518,"_type":366},{"_key":567,"_type":562,"oldSlug":568,"redirect":569},"clientfix-02","how-to-get-returns-from-ai-investments",{"_type":186,"page":570},{"_ref":7,"_type":366},{"_key":572,"_type":562,"oldSlug":573,"redirect":574},"clientfix-03","data-management-101",{"_type":186,"page":575},{"_ref":7,"_type":366},{"_createdAt":577,"_id":578,"_rev":579,"_system":580,"_type":583,"_updatedAt":584,"author":585,"category":654,"date":658,"dateModified":658,"description":659,"image":668,"name":678,"reviewer":679,"sections":747,"seo":1564,"slug":1567,"tags":189},"2026-08-20T09:21:09Z","ai-extract-precision-2026-w34","S3Sfff2lxcntH767szKs5K",{"base":581},{"id":578,"rev":582},"Q0lJJ1vxfjdElQtFKQa8ww","article","2026-08-20T09:34:18Z",{"_createdAt":586,"_id":587,"_rev":588,"_system":589,"_type":592,"_updatedAt":593,"alumniOf":594,"description":595,"image":612,"knowsAbout":615,"linkedin":616,"name":617,"orderRank":40,"position":618,"quote":619,"sections":620,"seo":648,"slug":652},"2026-02-17T11:28:47Z","6a828dd8-0fc0-47ad-a267-a3982dc7ae2f","4Hcc1NJ6s8BorOrDxpANE3",{"base":590},{"id":587,"rev":591},"KRroYNn1MtkjP4cuvgHzYQ","person","2026-08-07T15:13:09Z","The University of Edinburgh",[596,604],{"_key":597,"_type":337,"children":598,"markDefs":603,"style":355},"18f34a82adaf",[599],{"_key":600,"_type":341,"marks":601,"text":602},"e33d046985c0",[],"Ivan Angelinin is the CTO of Cosmos Thrace. He grew up in Bansko, a small mountain town in southwest Bulgaria, before moving abroad to study and work in technology. His path took him through Scotland and Switzerland, where he specialised in practical AI and machine learning, and then to one of Ireland's\nleading software companies, before returning home to join Cosmos Thrace, where he now leads the technical organisation. Inside the company he is\ndescribed as the backbone of the team. Externally, he is increasingly the public technical voice for Cosmos Thrace's positioning around Databricks, Unity\nCatalog, Medallion architecture, and the broader Data and AI agenda.",[],{"_key":605,"_type":337,"children":606,"markDefs":611,"style":355},"bbd915a2782a",[607],{"_key":608,"_type":341,"marks":609,"text":610},"045b9788961d",[],"As CTO, Ivan owns the firm's technical direction, sets the standard for how engagements are scoped and delivered, and runs the internal ML curriculum\nthat keeps the team current on Databricks and the wider Data and AI ecosystem. He is technical, direct, and known for being straight to the point, but his public commentary tends to do something different: it reframes engineering work in commercial terms. The pattern is captured in a line he is often quoted on: \"Your Databricks project isn't competing with the old system. It's competing against every other project in the company, fighting for the same budget.\" That ability to switch between technical and commercial registers in the same conversation is what makes Cosmos Thrace's engineering credible to CTOs and CFOs in the same room.",[],{"_type":82,"asset":613},{"_ref":614,"_type":366},"image-356175abf86a2f22e416ec5ec746e18711ea11a8-800x800-jpg","Databrikcs, Data Engineering, AI","https:\u002F\u002Fwww.linkedin.com\u002Fin\u002Fivan-angelinin-4001a689\u002F","Ivan Angelinin","Technical Director","CTO - The brain dude.",[621],{"_key":622,"_type":623,"afterTextLevel":624,"bigTextLevel":625,"button":626,"buttonPrimary":99,"certificates":627,"descriptionLevel":646,"panelTitle":647},"f7b36042f7be","certificatesSection","h4","h3",{"_type":186,"customName":99,"customUrl":99,"newTab":99},[628,634,640],{"_key":629,"image":630},"fccc5ee251ad",{"_type":143,"alt":631,"asset":632},"Databricks certification - Data Analyst Associate",{"_ref":633,"_type":366},"image-3b6722349836100f304df9f352ad5717390f1aa3-570x792-png",{"_key":635,"image":636},"dd039cb70a06",{"_type":143,"alt":637,"asset":638},"AWScertification - Data Analyst",{"_ref":639,"_type":366},"image-1ce2c177a01d042c7ccd069c883c9e2ded4e7601-100x100-png",{"_key":641,"image":642},"047653d6883a",{"_type":143,"alt":643,"asset":644},"Microsoft certification - Azure Data Engineer",{"_ref":645,"_type":366},"image-1764f8033e2078e858b9f348109b4926849b8593-100x100-png","large","certification",{"_type":649,"description":650,"title":651},"seo","Ivan Angelinin is Technical Director at Cosmos Thrace, leading Databricks implementations across EMEA with deep expertise in Unity Catalog and medallion patterns.","Ivan Angelinin - Technical Director at Cosmos Thrace",{"_type":17,"current":653},"ivan-angelinin",{"_createdAt":6,"_id":7,"_rev":8,"_system":655,"_type":12,"_updatedAt":13,"name":14,"orderRank":15,"slug":657,"uid":18},{"base":656},{"id":7,"rev":11},{"_type":17,"current":18},"2026-08-20",[660],{"_key":661,"_type":337,"children":662,"markDefs":667,"style":355},"d0",[663],{"_key":664,"_type":341,"marks":665,"text":666},"d0s0",[],"Databricks shipped Precision Mode for ai_extract this week, reporting 94.7% accuracy across roughly 9,000 complex documents and a seven-point margin over the strongest frontier-model baseline they tested. It is generally available now. The more useful question is not whether the benchmark is impressive. It is whether it changes the answer for the document-processing project your organisation shelved.",[],{"_type":143,"alt":669,"asset":670},"A deeply layered canyon at sunset with its far end lit orange, evoking the long documents Databricks AI Extract is built to read.",{"_id":671,"_type":87,"metadata":672,"url":677},"image-acb5e029068b7dbfa7bea5b1b1a9ffd00277e368-1456x816-png",{"dimensions":673},{"_type":95,"aspectRatio":674,"height":675,"width":676},1.7843137254901962,816,1456,"https:\u002F\u002Fcdn.sanity.io\u002Fimages\u002Fwn2c54m4\u002Fv2\u002Facb5e029068b7dbfa7bea5b1b1a9ffd00277e368-1456x816.png","Databricks AI Extract Precision Mode: What Changes for Document Processing Projects?",{"_createdAt":680,"_id":681,"_rev":682,"_system":683,"_type":592,"_updatedAt":79,"description":686,"image":736,"linkedin":739,"name":740,"orderRank":28,"position":741,"quote":741,"seo":742,"slug":745},"2026-02-17T11:28:19Z","54f6ac53-ca42-4f8f-8e5e-a8c9c73608b5","4Hcc1NJ6s8BorOrDxpABwg",{"base":684},{"id":681,"rev":685},"iG7LJCIXmUyigSDTCxVnCy",[687,706,713,721,729],{"_key":688,"_type":337,"children":689,"markDefs":703,"style":355},"b69d2ec948f7",[690,694,699],{"_key":691,"_type":341,"marks":692,"text":693},"a0e39ec7c52c",[],"Idan Harel is a serial founder, builder, and investor in enterprise technology. He is Managing Partner and Head of Business Development and Sales at\nCosmos Thrace, a Databricks Partner based in Sofia, Bulgaria. He founded ",{"_key":695,"_type":341,"marks":696,"text":698},"dbd48e5780e4",[697],"016ca837a4a6","Do IT Wise",{"_key":700,"_type":341,"marks":701,"text":702},"c71a6873a9a4",[],", a ServiceNow Elite Partner that grew to over 150 people\nbefore being acquired by Inetum in 2022, and Zigiwave, an integration product adopted by Fortune 500 companies. ",[704],{"_key":697,"_type":186,"href":705},"https:\u002F\u002Fwww.acta-verba.com\u002Fnews\u002Finetum-announces-the-strategic-acquisition-of-do-it-wise-a-servicenow-elite-partner\u002F",{"_key":707,"_type":337,"children":708,"markDefs":712,"style":355},"52183473a1f9",[709],{"_key":691,"_type":341,"marks":710,"text":711},[],"Beyond his operating roles, he is an active angel investor focused on Data and AI innovation, and is known across the regional tech community for combining commercial sharpness with contagious optimism.",[],{"_key":714,"_type":337,"children":715,"markDefs":720,"style":355},"190026a8a93b",[716],{"_key":717,"_type":341,"marks":718,"text":719},"6a1f8529edb5",[],"He joined Cosmos Thrace in 2025 with a clear mandate: sharpen the company's commercial direction and accelerate its evolution into Eastern Europe's most trusted partner for Databricks Consulting and Implementation and modern Data & AI solutions. The pivot toward a Databricks-first identity was led by him, and it has shaped the company's go-to-market, partnerships, and hiring decisions ever since. ",[],{"_key":722,"_type":337,"children":723,"markDefs":728,"style":355},"bbdf8bfefee5",[724],{"_key":725,"_type":341,"marks":726,"text":727},"11f77350a6d4",[],"",[],{"_key":730,"_type":337,"children":731,"markDefs":735,"style":355},"8a041834528a",[732],{"_key":717,"_type":341,"marks":733,"text":734},[],"He approaches every conversation through the lens of business outcome rather than technical purity, an orientation captured in a line he is often quoted on: \"If you can't tell the CFO how much money your project saves, it doesn't matter how well your Spark jobs run.\" That mindset, paired with what colleagues describe as a relentlessly positive energy and a habit of laughing his way through complex problems, has become a defining feature of how Cosmos Thrace shows up in the market.",[],{"_type":82,"asset":737},{"_ref":738,"_type":366},"image-a65e972e65887bb74d6c084e34c02491ec72178d-800x800-jpg","https:\u002F\u002Fwww.linkedin.com\u002Fin\u002Fidanharel\u002F","Idan Harel","Managing Partner",{"_type":649,"description":743,"title":744},"Idan Harel is Managing Partner at Cosmos Thrace, advising EMEA enterprises on Databricks ROI, partner selection, digital transformation, and AI strategy.","Idan Harel - Managing Partner at Cosmos Thrace",{"_type":17,"current":746},"idan-harel",[748,1460],{"_key":749,"_type":750,"afterTextLevel":624,"background":189,"bckg":189,"bigTextLevel":625,"button":751,"buttonPrimary":99,"cards":189,"certificates":189,"content":752,"descriptionLevel":646,"image":189,"images":189,"links":189,"redirect":189,"thumbnail":189},"body","richTextSection",{"_type":186,"customName":99,"customUrl":99,"newTab":99,"page":189},[753,762,777,789,809,821,833,845,857,869,877,885,903,911,919,927,968,980,992,1004,1012,1049,1065,1081,1093,1101,1109,1117,1125,1133,1141,1149,1157,1165,1173,1181,1189,1197,1205,1213,1221,1229,1237,1252,1260,1268,1276,1284,1292,1300,1317,1329,1337,1345,1357,1369,1381,1393,1405,1413,1422,1430,1445],{"_key":754,"_type":337,"children":755,"markDefs":760,"style":761},"b0002",[756],{"_key":757,"_type":341,"marks":758,"text":759},"s0001",[],"TL;DR",[],"h2",{"_key":763,"_type":337,"children":764,"level":774,"listItem":775,"markDefs":776,"style":355},"b0005",[765,770],{"_key":766,"_type":341,"marks":767,"text":769},"s0003",[768],"strong","Precision Mode is a new setting inside the existing ai_extract function",{"_key":771,"_type":341,"marks":772,"text":773},"s0004",[],", not a separate product. Set mode to precision, or toggle it in the Information Extraction UI.",1,"bullet",[],{"_key":778,"_type":337,"children":779,"level":774,"listItem":775,"markDefs":788,"style":355},"b0008",[780,784],{"_key":781,"_type":341,"marks":782,"text":783},"s0006",[768],"It is generally available.",{"_key":785,"_type":341,"marks":786,"text":787},"s0007",[]," Databricks did not publish pricing in the announcement.",[],{"_key":790,"_type":337,"children":791,"level":774,"listItem":775,"markDefs":808,"style":355},"b0013",[792,796,800,804],{"_key":793,"_type":341,"marks":794,"text":795},"s0009",[768],"Reported accuracy: 94.7%",{"_key":797,"_type":341,"marks":798,"text":799},"s0010",[]," across roughly 9,000 documents, ",{"_key":801,"_type":341,"marks":802,"text":803},"s0011",[768],"seven points ahead",{"_key":805,"_type":341,"marks":806,"text":807},"s0012",[]," of GPT-5.6 Sol, the strongest frontier-model chunk-and-merge baseline they benchmarked against.",[],{"_key":810,"_type":337,"children":811,"level":774,"listItem":775,"markDefs":820,"style":355},"b0016",[812,816],{"_key":813,"_type":341,"marks":814,"text":815},"s0014",[768],"Evaluated on ten internal datasets and five named public benchmarks:",{"_key":817,"_type":341,"marks":818,"text":819},"s0015",[]," VAREX, RealDocBench, LongExtractBench, LEDGER and the Caselaw Access Project.",[],{"_key":822,"_type":337,"children":823,"level":774,"listItem":775,"markDefs":832,"style":355},"b0019",[824,828],{"_key":825,"_type":341,"marks":826,"text":827},"s0017",[768],"It targets three specific failure modes",{"_key":829,"_type":341,"marks":830,"text":831},"s0018",[],", not extraction generally: documents too long for one pass, outputs too deeply nested to survive intact, and schemas that need reasoning rather than pattern matching.",[],{"_key":834,"_type":337,"children":835,"level":774,"listItem":775,"markDefs":844,"style":355},"b0022",[836,840],{"_key":837,"_type":341,"marks":838,"text":839},"s0020",[768],"Tested against documents up to 2,000 pages",{"_key":841,"_type":341,"marks":842,"text":843},"s0021",[]," and schemas with more than 300 deeply nested fields.",[],{"_key":846,"_type":337,"children":847,"level":774,"listItem":775,"markDefs":856,"style":355},"b0025",[848,852],{"_key":849,"_type":341,"marks":850,"text":851},"s0023",[768],"Named customers:",{"_key":853,"_type":341,"marks":854,"text":855},"s0024",[]," Panasonic, EY-Parthenon, Intercontinental Exchange.",[],{"_key":858,"_type":337,"children":859,"level":774,"listItem":775,"markDefs":868,"style":355},"b0028",[860,864],{"_key":861,"_type":341,"marks":862,"text":863},"s0026",[768],"Our take:",{"_key":865,"_type":341,"marks":866,"text":867},"s0027",[]," seven points is a real gain and an unglamorous one. It matters most to organisations holding a stalled business case, because the specific objection that stalled it may no longer be true.",[],{"_key":870,"_type":337,"children":871,"markDefs":876,"style":761},"b0030",[872],{"_key":873,"_type":341,"marks":874,"text":875},"s0029",[],"What was actually announced",[],{"_key":878,"_type":337,"children":879,"markDefs":884,"style":355},"b0032",[880],{"_key":881,"_type":341,"marks":882,"text":883},"s0031",[],"Precision Mode is a new mode within ai_extract, the document-extraction function already in the platform. It pairs two things:",[],{"_key":886,"_type":887,"caption":189,"headers":888,"rows":891},"tbl0033","table",[889,890],"Component","What it does",[892,898],{"_key":893,"_type":894,"cells":895},"r0034","row",[896,897],"Custom-trained extraction models","Models trained specifically for structured document extraction, rather than general-purpose LLMs pressed into the job",{"_key":899,"_type":894,"cells":900},"r0035",[901,902],"Agentic harness","Inspired by Databricks' MemEx work. It decomposes a large extraction job semantically, runs the subtasks in parallel, preserves intermediate results, and reconciles them into one structured output",{"_key":904,"_type":337,"children":905,"markDefs":910,"style":355},"b0037",[906],{"_key":907,"_type":341,"marks":908,"text":909},"s0036",[],"The second half is the interesting part architecturally. The standard approach to a document too large for a context window is chunk-and-merge: split it, extract from each piece, staple the results together. That is also the baseline Databricks benchmarked against, and the seven-point gap is the gap between stapling and reconciling.",[],{"_key":912,"_type":337,"children":913,"markDefs":918,"style":625},"b0039",[914],{"_key":915,"_type":341,"marks":916,"text":917},"s0038",[],"The three problems it targets",[],{"_key":920,"_type":337,"children":921,"markDefs":926,"style":355},"b0041",[922],{"_key":923,"_type":341,"marks":924,"text":925},"s0040",[],"Databricks is specific about the failure modes, and the specificity is worth reading closely, because it tells you whether your use case is in scope:",[],{"_key":928,"_type":82,"alt":929,"asset":930,"caption":967},"56ceff322fbc","The three ways document extraction breaks: too long to hold in one pass, too nested to survive intact, needs reasoning rather than pattern matching",{"_createdAt":931,"_id":932,"_rev":933,"_type":87,"_updatedAt":931,"assetId":934,"extension":90,"metadata":935,"mimeType":125,"originalFilename":962,"path":963,"sha1hash":934,"size":964,"uploadId":965,"url":966},"2026-08-20T09:33:38Z","image-08d50ed949038279cd5b03adfdb35ee2e3dc6bdf-1200x820-png","S3Sfff2lxcntH767szKjDG","08d50ed949038279cd5b03adfdb35ee2e3dc6bdf",{"_type":92,"blurHash":936,"dimensions":937,"hasAlpha":99,"isOpaque":100,"lqip":941,"palette":942,"thumbHash":961},"V2P?m-j^_3?b_2_3E1-pt7j[IUR%t6t3t79Fax%LNF%M",{"_type":95,"aspectRatio":938,"height":939,"width":940},1.4634146341463414,820,1200,"data:image\u002Fpng;base64,iVBORw0KGgoAAAANSUhEUgAAABQAAAAOCAIAAACgpqunAAAACXBIWXMAAAsTAAALEwEAmpwYAAAByUlEQVR4nFWTC5KrIBBF3f+qEiuTmYiKCArNH39kdvEKMHmZqlMUNn2Fvg3V09vD6t2oF++5js48vT1xH6M\u002FqQ6rDaMc94khjxmBeyCDoiQxZiiRIzET3bQ6xZuWrEVNXTf1tanrx\u002FXyc7k012uivqK6LvS3G75\u002FdV+3sXkE4Kd4N0qOhLaIooYilCZdy7p2yrDM3Hd5T2Ym6uZpN6+dozOLBMdnx2cv+KJg1XJVkJGF3ajD5vrdn7Kr6Oyq7KrMbvRhTXQmehudOayO+TPlBfcMLjp7GBP\u002FiL3btN91WiuhpH+Jz7zgfoOPzu1J\u002FL8F1WGNF8pzWCScJzR6lRBALCA2LQ+rD2dWrQKIAGJTsBtVflGtCgTGrG0FxkAGOzMv5hRBiCefBsenAGLGPb7fKWqA4Ldn1SLF1Lb453tsGta1MBI9Udai\u002Fvs+No8UIYOZGEXoUdf9\u002Fc5aBCNZQERnUp+BEIoQ61o+YM2oEzOQgXZo7js+YElHx2fF6NR3M+7FgBUdQxEfVnvBzcTsPAXgm05lBynyafmqoERWJQNwDzx5IZMXsfQ5ryW3ij3F7cLb\u002F92oRSZZSSs3v\u002FroTW7P52P4oOjzbdHvtH\u002FO0Mt3YL11DQAAAABJRU5ErkJggg==",{"_type":103,"darkMuted":943,"darkVibrant":946,"dominant":948,"lightMuted":949,"lightVibrant":952,"muted":955,"vibrant":958},{"_type":105,"background":944,"foreground":107,"population":945,"title":107},"#463d3f",1.32,{"_type":105,"background":947,"foreground":107,"population":115,"title":107},"#36275d",{"_type":105,"background":944,"foreground":107,"population":945,"title":107},{"_type":105,"background":950,"foreground":118,"population":951,"title":107},"#c3bcb4",0.13,{"_type":105,"background":953,"foreground":107,"population":954,"title":107},"#9880d3",0.02,{"_type":105,"background":956,"foreground":107,"population":957,"title":107},"#5d4982",0.04,{"_type":105,"background":959,"foreground":107,"population":960,"title":107},"#563e94",0.01,"NggCBYCmp6d2iYh2iDiIj2BWbvk2","fig2-failure-modes.png","images\u002Fwn2c54m4\u002Fv2\u002F08d50ed949038279cd5b03adfdb35ee2e3dc6bdf-1200x820.png",109540,"1e26bdb79fddbeedde9b0bb19268a14dd6ee95ec","https:\u002F\u002Fcdn.sanity.io\u002Fimages\u002Fwn2c54m4\u002Fv2\u002F08d50ed949038279cd5b03adfdb35ee2e3dc6bdf-1200x820.png","The three failure modes Databricks names. If none describe your documents, this release is not aimed at you.",{"_key":969,"_type":337,"children":970,"markDefs":979,"style":355},"b0044",[971,975],{"_key":972,"_type":341,"marks":973,"text":974},"s0042",[768],"1. Long documents needing cross-page reconciliation.",{"_key":976,"_type":341,"marks":977,"text":978},"s0043",[]," Their example: a lease where a term on page 1 depends on a clause on page 80. Chunking destroys that relationship, because neither chunk contains both facts.",[],{"_key":981,"_type":337,"children":982,"markDefs":991,"style":355},"b0047",[983,987],{"_key":984,"_type":341,"marks":985,"text":986},"s0045",[768],"2. Large nested outputs that get truncated.",{"_key":988,"_type":341,"marks":989,"text":990},"s0046",[]," An invoice with thousands of line items produces an output big enough that the tail is silently lost. Silently is the operative word.",[],{"_key":993,"_type":337,"children":994,"markDefs":1003,"style":355},"b0050",[995,999],{"_key":996,"_type":341,"marks":997,"text":998},"s0048",[768],"3. Complex schemas requiring reasoning.",{"_key":1000,"_type":341,"marks":1001,"text":1002},"s0049",[]," Synthesising across multiple documents, or applying conditional logic, rather than locating a field and copying it.",[],{"_key":1005,"_type":337,"children":1006,"markDefs":1011,"style":625},"b0052",[1007],{"_key":1008,"_type":341,"marks":1009,"text":1010},"s0051",[],"The benchmark",[],{"_key":1013,"_type":82,"alt":1014,"asset":1015,"caption":1048},"30f61560f307","Databricks AI Extract Precision Mode reported accuracy: 94.7 percent across roughly 9,000 complex documents, seven points ahead of the GPT-5.6 Sol chunk-and-merge baseline",{"_createdAt":1016,"_id":1017,"_rev":1018,"_type":87,"_updatedAt":1016,"assetId":1019,"extension":90,"metadata":1020,"mimeType":125,"originalFilename":1043,"path":1044,"sha1hash":1019,"size":1045,"uploadId":1046,"url":1047},"2026-08-20T09:32:19Z","image-e86d5e40697d835fcb3dc9c7abfc730ec54f773e-1200x675-png","Q0lJJ1vxfjdElQtFKQaqX0","e86d5e40697d835fcb3dc9c7abfc730ec54f773e",{"_type":92,"blurHash":1021,"dimensions":1022,"hasAlpha":99,"isOpaque":100,"lqip":1025,"palette":1026,"thumbHash":1042},"M4Q08Z00M}_2-:_2E0M{xu%M~o?ZbDs;s=",{"_type":95,"aspectRatio":1023,"height":1024,"width":940},1.7777777777777777,675,"data:image\u002Fpng;base64,iVBORw0KGgoAAAANSUhEUgAAABQAAAALCAIAAADwazoUAAAACXBIWXMAAAsTAAALEwEAmpwYAAABU0lEQVR4nH2S\u002FW6CMBTFef9H2Vxm5lziHOLHREBoqZV+0N4CzoKPsRTitj+myUmTJvd3z703x7vUpjX6rMte6qyVBTV8W6Mvtbkjr6sAiiNDGUMZx4hjJEmuKGkEt6C6Cu7BFhTehYvpdP4y8V+nW\u002F8Dhdsi3TeC3ycvA4y2n7PxePLw+DZ6ipeBoqQWzIK6T15q47VGswzFwTr0g2ixYimy2mFdBV0FrXHv7Z0NsKRI5hgtcxrSIi4a7lZtJBc5pvtEYHwq5Q1nALKhi1Gyek4zn+AV1USetZIkj4Kgv4Kvj\u002FRff8+NvT\u002Fs3qNwtsPrTKTO2Wqlj5TEUbrZHJKkEuwWDIYVLMuKNC0JaaS0oC2oRgrDi4qzUyktqNZoa5Q1uu019PK6ypx12UheC34qxZeS18C4kAx1XQVDjSv4kx+vP6wL2aDe9rf3j4YUWnD5s1fnb6SzMvMQ\u002FRnOAAAAAElFTkSuQmCC",{"_type":103,"darkMuted":1027,"darkVibrant":1029,"dominant":1031,"lightMuted":1034,"lightVibrant":1036,"muted":1039,"vibrant":1040},{"_type":105,"background":1028,"foreground":107,"population":951,"title":107},"#4c4443",{"_type":105,"background":1030,"foreground":107,"population":115,"title":107},"#34186b",{"_type":105,"background":1032,"foreground":107,"population":1033,"title":107},"#7755a8",0.4,{"_type":105,"background":1035,"foreground":118,"population":115,"title":107},"#bc9cc4",{"_type":105,"background":1037,"foreground":118,"population":1038,"title":107},"#a989da",0.11,{"_type":105,"background":1032,"foreground":107,"population":1033,"title":107},{"_type":105,"background":1041,"foreground":107,"population":960,"title":107},"#7444d4","NhgGBIBwqcxnlomsh5dQmf5rBg==","fig1-accuracy.png","images\u002Fwn2c54m4\u002Fv2\u002Fe86d5e40697d835fcb3dc9c7abfc730ec54f773e-1200x675.png",69133,"18f1fc6ce05240995c13f6b6c99aae8f3b041dbe","https:\u002F\u002Fcdn.sanity.io\u002Fimages\u002Fwn2c54m4\u002Fv2\u002Fe86d5e40697d835fcb3dc9c7abfc730ec54f773e-1200x675.png","Vendor-reported figures from a vendor-designed evaluation.",{"_key":1050,"_type":337,"children":1051,"markDefs":1064,"style":355},"b0056",[1052,1056,1060],{"_key":1053,"_type":341,"marks":1054,"text":1055},"s0053",[],"94.7% accuracy across roughly 9,000 documents, outperforming the strongest frontier-model chunk-and-merge baseline, ",{"_key":1057,"_type":341,"marks":1058,"text":1059},"s0054",[768],"GPT-5.6 Sol",{"_key":1061,"_type":341,"marks":1062,"text":1063},"s0055",[],", by seven points. Test documents ran up to 2,000 pages, with schemas exceeding 300 deeply nested fields and thousands of line items.",[],{"_key":1066,"_type":337,"children":1067,"markDefs":1080,"style":355},"b0060",[1068,1072,1076],{"_key":1069,"_type":341,"marks":1070,"text":1071},"s0057",[],"The evaluation covered ",{"_key":1073,"_type":341,"marks":1074,"text":1075},"s0058",[768],"ten internal datasets and five public benchmarks",{"_key":1077,"_type":341,"marks":1078,"text":1079},"s0059",[],". Databricks names them as:",[],{"_key":1082,"_type":337,"children":1083,"markDefs":1092,"style":355},"b0063",[1084,1088],{"_key":1085,"_type":341,"marks":1086,"text":1087},"s0061",[768],"Internal (10):",{"_key":1089,"_type":341,"marks":1090,"text":1091},"s0062",[]," datasets inspired by customer workloads across financial services, manufacturing and healthcare.",[],{"_key":1094,"_type":337,"children":1095,"markDefs":1100,"style":355},"b0065",[1096],{"_key":1097,"_type":341,"marks":1098,"text":1099},"s0064",[768],"Public (5):",[],{"_key":1102,"_type":337,"children":1103,"level":774,"listItem":775,"markDefs":1108,"style":355},"b0067",[1104],{"_key":1105,"_type":341,"marks":1106,"text":1107},"s0066",[],"VAREX",[],{"_key":1110,"_type":337,"children":1111,"level":774,"listItem":775,"markDefs":1116,"style":355},"b0069",[1112],{"_key":1113,"_type":341,"marks":1114,"text":1115},"s0068",[],"RealDocBench",[],{"_key":1118,"_type":337,"children":1119,"level":774,"listItem":775,"markDefs":1124,"style":355},"b0071",[1120],{"_key":1121,"_type":341,"marks":1122,"text":1123},"s0070",[],"LongExtractBench",[],{"_key":1126,"_type":337,"children":1127,"level":774,"listItem":775,"markDefs":1132,"style":355},"b0073",[1128],{"_key":1129,"_type":341,"marks":1130,"text":1131},"s0072",[],"LEDGER",[],{"_key":1134,"_type":337,"children":1135,"level":774,"listItem":775,"markDefs":1140,"style":355},"b0075",[1136],{"_key":1137,"_type":341,"marks":1138,"text":1139},"s0074",[],"Caselaw Access Project — used specifically for the long-document stress test",[],{"_key":1142,"_type":337,"children":1143,"markDefs":1148,"style":355},"b0077",[1144],{"_key":1145,"_type":341,"marks":1146,"text":1147},"s0076",[],"That list is worth noting for one reason: five named public benchmarks is more disclosure than most vendor accuracy claims come with, and it means the comparison is at least partly reproducible by someone who wants to check it.",[],{"_key":1150,"_type":337,"children":1151,"markDefs":1156,"style":355},"b0079",[1152],{"_key":1153,"_type":341,"marks":1154,"text":1155},"s0078",[],"Named early customers are Panasonic, EY-Parthenon and Intercontinental Exchange, with document types including 10-K filings, bills of lading, technical manuals, invoices, clinical notes and patent applications.",[],{"_key":1158,"_type":337,"children":1159,"markDefs":1164,"style":761},"b0081",[1160],{"_key":1161,"_type":341,"marks":1162,"text":1163},"s0080",[],"Why this matters for enterprises",[],{"_key":1166,"_type":337,"children":1167,"markDefs":1172,"style":625},"b0083",[1168],{"_key":1169,"_type":341,"marks":1170,"text":1171},"s0082",[],"1. Seven points is boring, and that is the point",[],{"_key":1174,"_type":337,"children":1175,"markDefs":1180,"style":355},"b0085",[1176],{"_key":1177,"_type":341,"marks":1178,"text":1179},"s0084",[],"Seven percentage points is not the kind of number that trends. It is, however, the kind of number that changes an operating model.",[],{"_key":1182,"_type":337,"children":1183,"markDefs":1188,"style":355},"b0087",[1184],{"_key":1185,"_type":341,"marks":1186,"text":1187},"s0086",[],"At 88% accuracy, a human checks everything, because you cannot know which 12% is wrong. At 95%, the conversation can move to exception handling: which extractions get flagged, who reviews them, what the fallback is. That is a different workflow with a different cost base, and it is the reason a marginal-sounding gain can unstall a project.",[],{"_key":1190,"_type":337,"children":1191,"markDefs":1196,"style":355},"b0089",[1192],{"_key":1193,"_type":341,"marks":1194,"text":1195},"s0088",[],"On a thousand documents, seven points is seventy fewer errors somebody has to find by hand. That arithmetic is ours, not Databricks', but it is the arithmetic a business case runs on.",[],{"_key":1198,"_type":337,"children":1199,"markDefs":1204,"style":625},"b0091",[1200],{"_key":1201,"_type":341,"marks":1202,"text":1203},"s0090",[],"2. The three failure modes are a scoping checklist",[],{"_key":1206,"_type":337,"children":1207,"markDefs":1212,"style":355},"b0093",[1208],{"_key":1209,"_type":341,"marks":1210,"text":1211},"s0092",[],"Most published guidance on document AI is about model choice. Databricks has instead named three structural reasons extraction breaks, and they map cleanly onto the kinds of documents enterprises actually hold.",[],{"_key":1214,"_type":337,"children":1215,"markDefs":1220,"style":355},"b0095",[1216],{"_key":1217,"_type":341,"marks":1218,"text":1219},"s0094",[],"If your problem is short, flat forms, this release changes little for you and simpler tooling was already sufficient. If your problem is 200-page contracts where obligations reference each other, or invoices whose line items outnumber the output budget, you were in the failing category and may no longer be.",[],{"_key":1222,"_type":337,"children":1223,"markDefs":1228,"style":355},"b0097",[1224],{"_key":1225,"_type":341,"marks":1226,"text":1227},"s0096",[],"That is a scoping question you can answer in an afternoon, without a vendor.",[],{"_key":1230,"_type":337,"children":1231,"markDefs":1236,"style":625},"b0099",[1232],{"_key":1233,"_type":341,"marks":1234,"text":1235},"s0098",[],"3. Extraction accuracy is not the same as extraction confidence",[],{"_key":1238,"_type":337,"children":1239,"markDefs":1251,"style":355},"b0103",[1240,1244,1248],{"_key":1241,"_type":341,"marks":1242,"text":1243},"s0100",[],"This is the gap worth watching. The announcement reports accuracy. It does not, in the material we have read, describe how the system signals ",{"_key":1245,"_type":341,"marks":1246,"text":1247},"s0101",[768],"when it is unsure",{"_key":1249,"_type":341,"marks":1250,"text":351},"s0102",[],[],{"_key":1253,"_type":337,"children":1254,"markDefs":1259,"style":355},"b0105",[1255],{"_key":1256,"_type":341,"marks":1257,"text":1258},"s0104",[],"That distinction decides the operating model. A system that is 95% accurate and silent about which 5% is wrong still needs full human review. A system that is 95% accurate and flags its own low-confidence extractions needs review of the flagged subset only. The second is dramatically cheaper to run, and it is the question we would put to Databricks before designing a workflow around this.",[],{"_key":1261,"_type":337,"children":1262,"markDefs":1267,"style":625},"b0107",[1263],{"_key":1264,"_type":341,"marks":1265,"text":1266},"s0106",[],"4. It runs where the data already is",[],{"_key":1269,"_type":337,"children":1270,"markDefs":1275,"style":355},"b0109",[1271],{"_key":1272,"_type":341,"marks":1273,"text":1274},"s0108",[],"ai_extract is callable from SQL and usable across the platform. For organisations already on Databricks, that removes the integration work, the second vendor and the data-movement question that usually accompanies a document-AI pilot. That is not a small operational advantage, and it is the strongest argument for evaluating this over a standalone extraction vendor.",[],{"_key":1277,"_type":337,"children":1278,"markDefs":1283,"style":761},"b0111",[1279],{"_key":1280,"_type":341,"marks":1281,"text":1282},"s0110",[],"The Cosmos Thrace perspective",[],{"_key":1285,"_type":337,"children":1286,"markDefs":1291,"style":355},"b0113",[1287],{"_key":1288,"_type":341,"marks":1289,"text":1290},"s0112",[],"We have not run Precision Mode in a client environment yet. This is a reading of the announcement, published while it is current, and we will say plainly when that changes.",[],{"_key":1293,"_type":337,"children":1294,"markDefs":1299,"style":355},"b0115",[1295],{"_key":1296,"_type":341,"marks":1297,"text":1298},"s0114",[],"What we can say from delivering data platform work across Europe is that document-extraction projects rarely die for the reason the post-mortem records. The recorded reason is usually accuracy. The actual reason is usually that nobody agreed what happens when the model is wrong.",[],{"_key":1301,"_type":337,"children":1302,"markDefs":1316,"style":355},"b0117",[1303,1307,1312],{"_key":1304,"_type":341,"marks":1305,"text":1306},"b47s0",[],"A 95% system and an 88% system fail differently in aggregate, but the ",{"_key":1308,"_type":341,"marks":1309,"text":1311},"b47s1",[1310],"em","design question they force is identical",{"_key":1313,"_type":341,"marks":1314,"text":1315},"b47s2",[],": who reviews, on what trigger, and what is the cost of a miss reaching a customer or a regulator. Teams that never answered that question will stall at 95% exactly as they stalled at 88%.",[],{"_key":1318,"_type":337,"children":1319,"markDefs":1328,"style":355},"b0120",[1320,1324],{"_key":1321,"_type":341,"marks":1322,"text":1323},"s0118",[],"So our honest advice is not \"the technology is ready now.\" It is narrower and more useful. ",{"_key":1325,"_type":341,"marks":1326,"text":1327},"s0119",[768],"If your business case named accuracy as the blocker, that line is out of date and the case deserves reopening. If your business case never had an exception-handling design, the new number will not save it.",[],{"_key":1330,"_type":337,"children":1331,"markDefs":1336,"style":355},"b0122",[1332],{"_key":1333,"_type":341,"marks":1334,"text":1335},"s0121",[],"One more note on the benchmark, in the spirit of how we read every vendor figure including our own: 94.7% is Databricks' number, about Databricks' product, on an evaluation Databricks designed. That does not make it wrong. It makes it a strong reason to run your own documents through it rather than a substitute for doing so.",[],{"_key":1338,"_type":337,"children":1339,"markDefs":1344,"style":761},"b0124",[1340],{"_key":1341,"_type":341,"marks":1342,"text":1343},"s0123",[],"What to do this month",[],{"_key":1346,"_type":337,"children":1347,"markDefs":1356,"style":355},"b0127",[1348,1352],{"_key":1349,"_type":341,"marks":1350,"text":1351},"s0125",[768],"1. Reopen the shelved business case.",{"_key":1353,"_type":341,"marks":1354,"text":1355},"s0126",[]," Find the sentence where somebody wrote that accuracy was not sufficient. That sentence may now be false. This costs an afternoon.",[],{"_key":1358,"_type":337,"children":1359,"markDefs":1368,"style":355},"b0130",[1360,1364],{"_key":1361,"_type":341,"marks":1362,"text":1363},"s0128",[768],"2. Check yourself against the three failure modes.",{"_key":1365,"_type":341,"marks":1366,"text":1367},"s0129",[]," Long documents with cross-page dependencies, deeply nested outputs, schemas needing reasoning. If none apply, this release is not aimed at you.",[],{"_key":1370,"_type":337,"children":1371,"markDefs":1380,"style":355},"b0133",[1372,1376],{"_key":1373,"_type":341,"marks":1374,"text":1375},"s0131",[768],"3. Run your own worst documents.",{"_key":1377,"_type":341,"marks":1378,"text":1379},"s0132",[]," Not the clean sample. The 300-page contract with scanned appendices that everybody avoids. Benchmarks are built on curated corpora; your estate is not curated.",[],{"_key":1382,"_type":337,"children":1383,"markDefs":1392,"style":355},"b0136",[1384,1388],{"_key":1385,"_type":341,"marks":1386,"text":1387},"s0134",[768],"4. Ask the confidence question.",{"_key":1389,"_type":341,"marks":1390,"text":1391},"s0135",[]," How does the system indicate uncertainty, and can that signal drive a review queue? Design the workflow around the answer, not around the headline number.",[],{"_key":1394,"_type":337,"children":1395,"markDefs":1404,"style":355},"b0139",[1396,1400],{"_key":1397,"_type":341,"marks":1398,"text":1399},"s0137",[768],"5. Price it before you plan it.",{"_key":1401,"_type":341,"marks":1402,"text":1403},"s0138",[]," Pricing was not stated in the announcement. Get it confirmed before it lands in a business case.",[],{"_key":1406,"_type":337,"children":1407,"markDefs":1412,"style":761},"b0165",[1408],{"_key":1409,"_type":341,"marks":1410,"text":1411},"s0164",[],"Sources",[],{"_key":1414,"_type":337,"children":1415,"level":774,"listItem":1420,"markDefs":1421,"style":355},"b0167",[1416],{"_key":1417,"_type":341,"marks":1418,"text":1419},"s0166",[],"Databricks — \"Databricks Document Intelligence: Pushing the Frontier of Complex Document Extraction\": https:\u002F\u002Fwww.databricks.com\u002Fblog\u002Fdatabricks-document-intelligence-pushing-frontier-complex-document-extraction","number",[],{"_key":1423,"_type":337,"children":1424,"markDefs":1429,"style":761},"b0169",[1425],{"_key":1426,"_type":341,"marks":1427,"text":1428},"s0168",[],"Related reading",[],{"_key":1431,"_type":337,"children":1432,"level":774,"listItem":775,"markDefs":1442,"style":355},"b0173",[1433,1438],{"_key":1434,"_type":341,"marks":1435,"text":1437},"s0171",[1436],"lnk0170","Databricks Lakebridge: What It Does, and What It Doesn't",{"_key":1439,"_type":341,"marks":1440,"text":1441},"s0172",[]," — the same argument applied to migration tooling: what the tool now does for free, and what it leaves you",[1443],{"_key":1436,"_type":186,"href":1444},"https:\u002F\u002Fcosmosthrace.com\u002Fresources\u002Fdatabricks\u002Fdatabricks-lakebridge-what-it-does-and-doesnt",{"_key":1446,"_type":337,"children":1447,"level":774,"listItem":775,"markDefs":1457,"style":355},"b0175",[1448,1452],{"_key":1449,"_type":341,"marks":1450,"text":1451},"b60s0",[],"Databricks migration services hub",{"_key":1453,"_type":341,"marks":1454,"text":1456},"b60s1",[1455],"md60_0"," — how we scope and run a migration, by source system",[1458],{"_key":1455,"_type":186,"href":1459},"https:\u002F\u002Fcosmosthrace.com\u002Fdatabricks-migration",{"_key":1461,"_type":1462,"accordion":1463,"afterTextLevel":624,"background":189,"bckg":189,"bigText":1561,"bigTextLevel":625,"button":1562,"buttonPrimary":99,"cards":189,"certificates":189,"content":189,"descriptionLevel":646,"image":189,"images":189,"links":189,"panelTitle":1563,"redirect":189,"thumbnail":189},"faq","accordionSection",{"items":1464},[1465,1477,1489,1501,1513,1525,1537,1549],{"_key":1466,"description":1467,"title":1476},"faq1",[1468],{"_key":1469,"_type":337,"children":1470,"markDefs":1475,"style":355},"f1b",[1471],{"_key":1472,"_type":341,"marks":1473,"text":1474},"f1s",[],"A mode within the existing ai_extract function, aimed at complex document extraction. It combines custom-trained extraction models with an agentic harness that decomposes large jobs, runs subtasks in parallel and reconciles them into one structured output.",[],"What is Databricks AI Extract Precision Mode?",{"_key":1478,"description":1479,"title":1488},"faq2",[1480],{"_key":1481,"_type":337,"children":1482,"markDefs":1487,"style":355},"f2b",[1483],{"_key":1484,"_type":341,"marks":1485,"text":1486},"f2s",[],"Databricks reports 94.7% accuracy across roughly 9,000 complex documents, seven points ahead of GPT-5.6 Sol, the strongest frontier-model chunk-and-merge baseline in their evaluation. The evaluation used ten internal datasets and five public benchmarks: VAREX, RealDocBench, LongExtractBench, LEDGER and the Caselaw Access Project. This is a vendor figure from a vendor-designed evaluation.",[],"How accurate is Precision Mode?",{"_key":1490,"description":1491,"title":1500},"faq3",[1492],{"_key":1493,"_type":337,"children":1494,"markDefs":1499,"style":355},"f3b",[1495],{"_key":1496,"_type":341,"marks":1497,"text":1498},"f3s",[],"Yes. Set the mode parameter to precision in ai_extract, or toggle it in the Information Extraction UI.",[],"Is Precision Mode generally available?",{"_key":1502,"description":1503,"title":1512},"faq4",[1504],{"_key":1505,"_type":337,"children":1506,"markDefs":1511,"style":355},"f4b",[1507],{"_key":1508,"_type":341,"marks":1509,"text":1510},"f4s",[],"Databricks did not state pricing in the announcement. Confirm it before building a business case.",[],"How much does it cost?",{"_key":1514,"description":1515,"title":1524},"faq5",[1516],{"_key":1517,"_type":337,"children":1518,"markDefs":1523,"style":355},"f5b",[1519],{"_key":1520,"_type":341,"marks":1521,"text":1522},"f5s",[],"Complex ones: long documents with cross-page dependencies, deeply nested outputs such as invoices with thousands of line items, and schemas requiring reasoning. Named document types include 10-K filings, bills of lading, technical manuals, invoices, clinical notes and patent applications.",[],"What kinds of documents is it for?",{"_key":1526,"description":1527,"title":1536},"faq6",[1528],{"_key":1529,"_type":337,"children":1530,"markDefs":1535,"style":355},"f6b",[1531],{"_key":1532,"_type":341,"marks":1533,"text":1534},"f6s",[],"Databricks reports testing on documents up to 2,000 pages and schemas with more than 300 deeply nested fields.",[],"How large a document can it handle?",{"_key":1538,"description":1539,"title":1548},"faq7",[1540],{"_key":1541,"_type":337,"children":1542,"markDefs":1547,"style":355},"f7b",[1543],{"_key":1544,"_type":341,"marks":1545,"text":1546},"f7s",[],"For extraction itself, less than you did. The work that remains is scoping which documents are worth processing, designing exception handling for the cases the model gets wrong, and integrating the output into a process people trust. None of that is a function call.",[],"Do I still need a partner if extraction is now a SQL function?",{"_key":1550,"description":1551,"title":1560},"faq8",[1552],{"_key":1553,"_type":337,"children":1554,"markDefs":1559,"style":355},"f8b",[1555],{"_key":1556,"_type":341,"marks":1557,"text":1558},"f8s",[],"The announcement reports accuracy rather than confidence signalling. We would treat this as an open question and confirm it with Databricks before designing a review workflow.",[],"Does it tell me when it is unsure?","What people ask about this topic",{"_type":186,"customName":99,"customUrl":99,"newTab":99,"page":189},"FAQ",{"_type":649,"description":1565,"ogImage":189,"title":1566},"Databricks reports 94.7% accuracy on complex document extraction, seven points ahead of the strongest frontier baseline. What it changes for a shelved project.","Databricks AI Extract Precision Mode: What Changes",{"_type":17,"current":1568},"databricks-ai-extract-precision-mode"]